Sense
Capture real-time motor condition data through industrial sensing, including vibration, electrical measurements, temperature, and other operating signals.
TwinSight
An end-to-end predictive maintenance system combining sensing, signal processing, machine learning, remaining useful life estimation, and a Unity-based Digital Twin.
The Problem
Industrial motors are critical assets, yet maintenance decisions are often reactive or based on fixed schedules. This can allow developing faults to go unnoticed until they become costly, while unnecessary scheduled maintenance can lead to premature component replacement. The challenge was to create a monitoring workflow that could turn real-time motor data into meaningful information about machine condition and support earlier, more informed maintenance decisions.
The Approach
TwinSight connects sensing, signal processing, machine learning, and Digital Twin visualization in one continuous predictive-maintenance workflow.
Process
Transform raw signals using FFT, Welch spectral analysis, Hilbert envelope analysis, overlapping windows, and engineered condition-monitoring features.
Predict
Use machine learning to classify motor condition across Nominal, Unbalance, Misalignment, and Combined fault states, while supporting remaining useful life estimation.
Visualize
Present live machine condition, signal behavior, fault states, and what-if scenarios through a Unity-based Digital Twin connected to the monitoring system.
Technical Decisions
Several modeling and validation choices were made to build a more reliable machine-learning pipeline from real motor recordings.
Grouped Validation
Grouped validation was used to reduce recording-level leakage and provide a more realistic estimate of model performance.
Robust Feature Scaling
RobustScaler was used to reduce sensitivity to extreme feature values while preserving the information needed for classification.
Structured Model Tuning
GridSearchCV was used to systematically explore model configurations and select stronger hyperparameter settings.
Results
MECHANICAL FAULT CLASSIFICATION
Reliable classification across four motor conditions
The final classifier distinguishes between Nominal, Unbalance, Misalignment, and Combined operating conditions, demonstrating the value of combining engineered signal features with machine learning for motor condition monitoring.
Why It Matters
TwinSight brings sensing, signal processing, machine learning, remaining useful life estimation, and Digital Twin visualization into one connected workflow. Rather than treating fault classification as an isolated model, the project demonstrates how AI can become part of a broader industrial monitoring system that turns physical machine signals into information that can support maintenance decisions.
PROJECT IN ACTION
System Demonstrations
The system was demonstrated across different motor operating conditions, showing how the physical setup, monitoring interface, and Digital Twin respond to changes in machine condition.
NORMAL OPERATION
Nominal Condition
Demonstration of the monitoring system during normal motor operation.
FAULT CONDITION
Unbalance
Demonstration of system behavior under an unbalance condition.
FAULT CONDITION
Misalignment
Demonstration of system behavior under a misalignment condition.
Need a Similar Solution?
If your project involves data, machine learning, forecasting, computer vision, NLP, or intelligent engineering systems, let’s discuss the problem and the right technical approach.